Variability in gene expression underlies incomplete penetrance

Abstract

The phenotypic differences between individual organisms can often be ascribed to underlying genetic and environmental variation. However, even genetically identical organisms in homogenous environments vary, suggesting that randomness in developmental processes such as gene expression may also generate diversity. In order to examine the consequences of gene expression variability in multicellular organisms, we studied intestinal specification in the roundworm Caenorhabditis elegans in which wild-type cell fate is invariant and controlled by a small transcriptional network. Mutations in elements of this network can have indeterminate effects: some mutant embryos fail to develop intestinal cells, while others produce intestinal precursors. By counting transcripts of the genes in this network in individual embryos, we show that the expression of an otherwise redundant gene becomes highly variable in the mutants and that this variation is thresholded to produce an ON/OFF expression pattern of the master regulatory gene of intestinal differentiation. Our results demonstrate that mutations in developmental networks can expose otherwise buffered stochastic variability in gene expression, leading to pronounced phenotypic variation.

In 1925, Timoféeff-Ressovsky and Romaschoff independently noticed that individuals harboring identical mutant alleles often exhibit either mutant or wild-type phenotypes, a property known as incomplete penetrance of the mutant phenotype^1–5. Such variation can sometimes be traced to differences in genetic background or environmental conditions, but can also arise from random fluctuations in processes such as gene expression⁶. In clonal populations of microbes, these stochastic effects in gene expression can be used as a mechanism for generating phenotypic variation^7–10, and multicellular organisms can also use variability to generate different cell types^11–13. In general, though, the gene expression patterns of different cells during metazoan development must be coordinated to ensure proper tissue formation. This suggests that stochastic fluctuations in gene expression may be controlled or their effects may be buffered under normal conditions. Here, we examined the consequences of random variability in gene expression during intestinal specification by measuring expression in individual C. elegans embryos using a fluorescence in situ hybridization technique capable of detecting single mRNA molecules. We found that expression in the wild type network was highly regular, but that mutations to components of this network, that are incompletely penetrant for loss of intestinal cells, led to large variations in the expression of a downstream gene. These variations were subsequently thresholded to yield alternative cell fates, showing that incomplete penetrance can result from stochastic fluctuations in gene expression when mutations compromise mechanisms that normally buffer such variability

Incomplete penetrance in intestinal development

The C. elegans intestine consists of 20 cells descended entirely from the E cell, which arises early in embryonic development (Fig. 1a). Intestinal cell-fate specification results from the activity of a short transcriptional cascade, beginning with the maternal deposition of skn-1 transcripts and ending with the expression of elt-2, which then activates hundreds of other genes during intestinal differentiation^15–17 (Fig. 1b). Genetic analyses have shown that skn-1 activates the genes med-1 and med-2 (which are essentially identical)^18,19, end-3 and end-1¹⁹. END-3 activates end-1¹⁹, and both gene products act in an OR-like fashion to activate elt-2^20–22, which maintains its own expression through a positive feedback loop^15,16,23. We examined whether the developmental consequences of mutations to skn-1 resulted from variability in the expression of genes in this pathway.

Figure 1. Gene expression in the C. elegans intestinal cell fate specification network.

a. The early embryonic lineage leading to the formation of the E cell. b. The gene regulatory network governing intestinal cell specification. skn-1 and pop-1 transcripts are maternally deposited. c–f. Visualization of single transcripts in individual wild-type (N2) embryos with DAPI as a nuclear counterstain. For Cy5, we assigned transcripts to med-1,2 in embryos with less than 30 nuclei and to elt-2 in those with more than 30 nuclei (f). Expression of elt-2 in g. wild-type (N2) and h. mutant embryos harboring the skn-1(zu129) allele.

We counted the number of mRNAs transcribed from these genes in individual embryos by using a version of fluorescence in situ hybridization that renders each mRNA visible as a single diffraction-limited fluorescent spot¹⁴ (Fig. 1d–f). Co-staining with DAPI allowed us to count the number of nuclei in each embryo, enabling stage-specific measurement of gene expression (Fig. 1c). We observed that all wild-type embryos beyond the 65-cell stage (when there are 4 cells in the E lineage) contained large numbers of elt-2 transcripts (Fig. 1g). skn-1 mutant embryos, however, die in late embryogenesis with most but not all embryos lacking intestinal cells²⁴. elt-2 expression in skn-1 mutant embryos was bimodal, with elt-2 exhibiting an ON/OFF expression pattern¹⁵ (Fig. 1h).

Intestinal network gene expression dynamics

To study the sources of this variability, we counted transcripts and nuclei in hundreds of differently staged mutant and wild-type embryos, thereby reconstructing the dynamics of the intestinal specification network (Fig. 2a, Supplementary Fig. 1). In the wild type, med-1,2 were the first zygotically expressed genes, followed by end-3, then end-1, then elt-2, in concordance with previous studies^25,26. We found no evidence for maternal med-1/2 transcripts^19,27, but small numbers of end-3 and end-1 transcripts were uniformly distributed throughout one-cell and two-cell embryos indicating that these originated in the mother’s gonad²⁸ (Supplementary Fig. 2). During periods of peak expression, transcript levels were similar for all these genes and did not display high variability.

Figure 2. Expression dynamics in wild-type and skn-1 mutant embryos.

a. Transcript number vs. number of nuclei for a collection of randomly staged wild-type (N2) (left) and zu135 mutant (right) early embryos. b. Depiction of the operation of the gut differentiation network in skn-1 mutant embryos. c. Number of cells expressing end-1 (top) or elt-2 (bottom) within individual wild-type and zu135 mutant embryos vs. number of nuclei. d. Transcript number vs. number of cells expressing end-1 (top) or elt-2 (bottom) in zu135 mutant embryos.

In contrast, the expression of these genes was far more variable in embryos homozygous for the skn-1 alleles zu67, zu129, or zu135 (Supplementary Fig. 3). The expression of elt-2 was bimodal; both the zu67 and zu135 alleles were more penetrant than zu129 for loss of elt-2 expression, in agreement with published morphological results²⁴. This difference in penetrance was robust to the choice of threshold dividing OFF from ON expression (Supplementary Fig. 4).

The skn-1 mutations also affected expression patterns of the genes upstream of elt-2 in several ways. med-1,2 transcripts were essentially absent, and end-3 transcript numbers were greatly diminished (Fig. 2a and Supplementary Fig. 3), effectively removing the med-1,2 and end-3 nodes of the gene network. This compromised end-1 activation, and left it as the sole activator of elt-2 (Fig. 2b). end-1 expression changed in two respects. First, end-1 expression began approximately one cell cycle later than in the wild type (Supplementary Fig. 1). Second, inter-embryo variation in end-1 expression was much higher than in wild-type worms (coefficient of variation 0.68, 0.60, 0.68, versus 0.16 for zu129, zu135, zu67, and N2, respectively; Supplementary Fig. 5, 6).

The number of cells expressing end-1 and elt-2 within individual mutant embryos also varied greatly from embryo to embryo (Fig. 2c and Supplementary Fig. 3). Furthermore, the total level of elt-2 expression correlated well with the total number of expressing cells. This suggests that each cell expressing elt-2 produced a constant amount of transcript (around 20–30% lower than the wild type; see Supplementary Fig. 7) and that expression variability came largely from differences in the number of actively transcribing cells (Fig. 2d, bottom, Supplementary Fig. 7), consistent with the bimodal expression pattern one would expect from a gene exhibiting self-activation. The expression of end-1, however, was quite variable even for a given total number of end-1 expressing cells (Fig. 2d, top; Supplementary Fig. 7) and even between cells in the same embryos (Supplementary Fig. 8).

end-1 must reach a threshold to activate elt-2

Since elt-2 activation in these mutant embryos depended primarily if not solely on end-1, we hypothesized that end-1 expression needed to reach a threshold level during a critical developmental time window in order to activate elt-2. Sub-threshold levels of end-1 would fail to induce elt-2 expression (Fig. 3a). In order to test this hypothesis, we looked for a relationship between levels of end-1 and activation of elt-2 in skn-1 mutant embryos (Fig. 3b). For all skn-1 mutants, elt-2 expression was only found in embryos with high levels of end-1 expression between the 65-cell and 120-cell, whereas both genes were highly expressed in wild-type embryos during this time. The same pattern was apparent in individual cells (Supplementary Fig. 9). After the 120-cell stage, most mutant embryos had negligible levels of elt-2 expression (Fig. 3b, right panels). In embryos that did express elt-2 highly, distributions of end-1 and elt-2 were similar to wild-type. This is consistent with our hypothesis that end-1 expression needs to reach a threshold, perhaps during a critical developmental window, in order to activate elt-2. The threshold itself may be caused by the self-activation of elt-2, with a certain amount of expression (modulated by END-1) being required to trigger the feedback loop⁸.

Figure 3. High levels of end-1 are required for elt-2 expression in skn-1 mutant embryos.

a. Model in which end-1 expression must surpass a threshold during a window of developmental time in order to activate elt-2 expression. b. Scatter plots of end-1 and elt-2 transcript numbers in wild-type (N2; blue) and skn-1 mutant embryos (red). c. Transcript number vs. number of cells expressing end-1 in zu67 mutant embryos containing between 65 and 120 nuclei. d. Number of elt-2 vs. end-1 transcripts in zu67 mutant embryos (c) with 1 through 8 (top to bottom) cells expressing end-1.

We also examined how the number of cells expressing end-1 influenced the decision to express elt-2. We found that only embryos expressing end-1 in four or more cells expressed enough end-1 to activate elt-2 (Fig. 3c–d, Supplementary Figure 10). An analysis of the dynamics of intercellular variability in cells expressing end-1 indicated that the thresholding decision was made when there are two E cells (Supplementary Fig. 11).

The zu67 and zu135 alleles of skn-1 were more penetrant for lack of elt-2 expression than the zu129 allele. Between the 65-cell and 120-cell stages, the distributions of end-1 transcript number in the zu67, zu129, and zu135 strains were not significantly different (p = 0.11, Anderson-Darling test). However, the thresholds differed between strains, with elt-2 activation requiring 259 and 249 end-1 transcripts in zu67 and zu135 embryos compared to 143 in zu129 embryos (Fig. 3b). This indicates that the lower penetrance in zu129 was primarily the result of a lowered threshold (Supplementary Fig. 12).

The greatly increased variability in end-1 expression in skn-1 mutants could originate from several sources, including transmitted variability in skn-1 or end-3 expression or fluctuations in the effectiveness of a POP-1 mediated Wnt signal important for activating end-1^29–31 (Fig. 1b). Analysis of skn-1 and end-3 mRNA levels in mutant strains argues against the first two possibilities (Supplementary Figs. 6, 13, 14, 15). pop-1 mRNA levels in the skn-1 mutant embryos were virtually identical to those in the wild-type (Supplementary Fig. 16), and POP-1 protein localization in the skn-1 mutants was similar to that observed in the wild type (data not shown), showing that fluctuations in pop-1 expression are unlikely to play a significant role in end-1 expression variability.

Chromatin remodeling affects variability

Several recent studies have implicated fluctuations in chromatin state between transcriptionally active and inactive conformations as a major source of variability in gene expression^32–34. In the intestinal specification network, end-1 is maintained in a transcriptionally inactive state through the activity of the histone deacetyltransferase HDA-1³⁰. skn-1 activates end-1 by recruiting the p300/CBP homolog CBP-1, which remodels the chromatin into a transcriptionally active state^35,36. In the skn-1 mutants, inefficient recruitment of CBP-1 to the end-1 promoter could result in more sporadic transcriptional activation, leading to increased expression variability^32,37–39. Thus, if the transcriptional repression by hda-1 were relieved, this variability should decrease because the activation of end-1 would no longer depend on the intermittent activity of CBP-1. To test this hypothesis, we measured intestinal network expression variability in skn-1 mutant embryos in which we downregulated hda-1 expression by RNAi. We found that elt-2 expression levels increased greatly, with virtually every embryo past the 100 cell stage showing some degree of elt-2 expression (Fig. 4a–b, Supplementary Fig. 17). This increase was due to a reduction in end-1 expression variability and a shift in the expression distribution towards wild-type levels (Fig. 4c).

Figure 4. Chromatin regulators and indirect network connections regulate variability in end-1 expression.

a. Expression dynamics in the zu135 strain subjected to RNAi against hda-1. b. Depiction of the role of hda-1 in the gut differentiation network. c. Histograms of the number of end-1 transcripts in wild-type (top; coefficient of variation of 0.20±0.057; error obtained by bootstrapping), skn-1(zu135) (middle; cv of 0.69±0.066) and skn-1(zu135); hda-1(RNAi) (bottom; cv of 0.44±0.056) embryos containing between 45 and 75 nuclei. d. Expression dynamics with an end-3 deletion. e. Depiction of the gut differentiation network with end-3 deleted.

HDA-1 is a global regulator and could a priori increase elt-2 expression in a manner independent of the increase in end-1 expression. We think this is not the case for two reasons. First, end-1 and elt-2 expression are still correlated in these embryos (Supplementary Fig. 18), strongly suggesting that the increase in elt-2 expression is directly related to the increase in end-1 expression. Second, the expression of med-1,2 and end-3 remained low, suggesting that the removal of hda-1 specifically affected end-1. These data imply that the proper regulation of chromatin may play a major role in controlling variability in gene expression.

Redundancy controls expression variability

Redundancy is a prominent feature of the endoderm specification network²², and may play a role in controlling developmental errors⁴⁰. All three skn-1 mutations affect the expression of end-1 both directly by hindering the transcriptional activation of end-1 by the SKN-1 protein and indirectly by downregulating med-1,2 and end-3^19,41,42. To measure how much end-1 expression variability was caused by the lack of end-3 expression, we measured transcript numbers in a strain with an end-3 deletion (ok1448) ²²(Fig. 4b–c). end-1 expression was not delayed as it was in the skn-1 mutants (Supplementary Figure 1), suggesting that end-3 is not important for the initiation of end-1 expression. We did, however, find that end-3 mutants occasionally displayed low levels of end-1 expression between the 40-cell and 120-cell stages, showing that some of the end-1 expression variability in the skn-1 mutants stems from a lack of end-3 expression.

END-3 also acts in concert with END-1 in a largely redundant fashion to activate elt-2²². To investigate this regulation, we measured elt-2 expression in end-1(ok558) and end-3(ok1448) deletion mutants. Deletion of end-1 had no discernible effects on elt-2 expression (Supplementary Fig. 19). end-3 mutants, however, exhibited variable delays in the activation of elt-2, although eventually, almost all embryos expressed appreciable (yet variable) levels of elt-2 (consistent with the roughly 5% of end-3 knockout animals that lack intestinal cells²²). This shows that some of the variability in elt-2 expression in the skn-1 mutants was due to the lack of appreciable end-3 expression in those embryos and helps explain why some skn-1 mutant embryos with high levels of end-1 failed to express elt-2. As in the skn-1 mutants, end-3 mutant embryos with low levels of end-1 expression exhibited low levels of elt-2 expression (Supplementary Fig. 18). However, many more embryos had delays in elt-2 expression than had low levels of end-1. While end-1 alone is not very efficient at activating elt-2 expression at precisely the right time, it is able to activate elt-2 eventually if present in sufficient quantity.

Theoretical^43,44 and experimental^45,46 work suggests that connections between different genes in a regulatory network may buffer genetic and environmental variation. In particular, removing more highly connected genes results in a greater susceptibility to variation than less connected genes. While the intestinal differentiation network is small with many unknown interactions, our data is at least consistent with the notion that connectivity may also help buffer stochastic variability. In the intestinal specification network, end-1 regulates the smallest number of genes while end-3 and skn-1 are progressively more connected, suggesting that the amount of gene expression variability induced corresponds to the degree of connectivity of the removed node.

Concluding remarks

We have demonstrated that the incomplete penetrance of the skn-1 mutant phenotype is a consequence of large variations in gene expression that are thresholded during development to determine cell fate¹. Our single molecule methodology allowed us to quantitatively measure these effects, showing that mutations to genes in the network alter the topology and compromise the logic of the intestinal specification network, leading to changes in gene expression levels, variability, and timing. In particular, the high variability in the mutants shows that metazoan gene expression can be highly variable and that wild-type metazoan developmental networks control those fluctuations⁴⁷. While inhibition of global regulators like Hsp90 can disrupt this buffering to expose hidden genetic, environmental and stochastic variation^48,49, we show that variability also arises from mutations to genes with far more specific functions, such as skn-1 and end-3. Thus, random variability may play a role in driving the evolution of buffering mechanisms⁵⁰. We propose that stochastic fluctuations in gene expression may underlie the phenotypic variation that often arises in mutant organisms even in fixed genetic and environmental contexts and anticipate that studies like ours may help elucidate the features of developmental networks that control the effects of underlying variation.

Methods summary

We harvested and fixed embryos from synchronized cultures of wild-type (N2) and mutant nematodes grown at 25°C. We maintained the skn-1 mutant alleles as heterozygotes using a balancer containing a fluorescent protein reporter; we isolated homozygotic embryos by sorting based on fluorescence. We performed FISH on the embryos and counted the mRNAs as described in Raj et al.¹⁴. We manually processed the images to determine the location and number of nuclei in each embryo using custom software written in MATLAB.

Online Materials and Methods

Strain construction

We replaced the nT1 [Unc n754] translocation in the skn-1 mutants skn-1(zu67), skn-1(zu129) and skn-1(zu135) in strains EU1, EU40 and EU31, respectively¹¹ with an nT1 translocation carrying the dominant marker qls51 [myo-2::gfp] and maintained the stocks as GFP positive skn-1/nT1[qIs51] heterozygotes. We also used the strains VC271 and RB1331 with genotypes end-1(ok558) and end-3(ok1448), respectively. Both are deletion alleles from the C. elegans Knockout Consortium.

Worm growth, sorting and fixation

To grow large quantities of synchronized worms, we harvested embryos by bleaching gravid adults, washing the embryos twice with water and then resuspending the worms in S-medium. These embryos hatched and then undergo growth arrest at the L1–L2 transition. We then plated these synchronized worms and grew them at 25°C. For the wild-type (N2), end-1(ok558) and end-3(ok1448) strains, we grew the worms until almost all the hermaphrodites were gravid. At this point, many embryos had already been laid; to remove those later staged embryos, we ran the culture through a 40 micron cell strainer (Becton Dickinson), which retains the gravid adults while allowing the free embryos to pass through. At this point, we fixed the embryos as described in ref. 19 using solutions kept at 25°C to minimize temperature variations, after which we performed fluorescence in situ hybridization on the embryos as described below.

For the skn-1 mutant strains (zu67, zu135, zu129 balanced by the translocation nT1[qIs51] which contains a pharyngeal GFP marker), we needed to isolate large numbers of embryos that were homozygous for the mutant allele of skn-1. Since skn-1 is a maternally deposited mRNA, these adults (which are GFP-negative) are viable and the effects of the skn-1 mutations show up in their offspring. To collect skn-1 homozygotes, we first grew a synchronized culture for two days at 25°C, after which we ran the worms through a Union Biometrica BioSorter to isolate only the GFP-negative worms. This sorting procedure resulted in fewer than 1 in 500 of the resulting population being GFP positive. After sorting, we placed the homozygous mutant worms back at 25°C for 24 hours, at which point all the worms were gravid. We then fixed them as outlined above. To perform the hda-1 RNAi on zu135 mutant embryos, we sorted for zu135 homozygotes using the BioSorter and then moved the worms onto plates seeded with bacteria expressing double-stranded RNA corresponding to hda-1 once they reached the L4 stage, after which we fixed the resultant embryos as described. We also grew some zu135 homozygotes on plates seeded with control bacteria (HT115(DE3); L4440 empty vector), finding expression patterns similar to those found on plates seeded with OP50.

Fluorescent in situ hybridization and imaging

We performed fluorescence in situ hybridization (FISH) as outlined in ref. 16. All hybridizations were performed in solution using probes coupled to either tetramethylrhodamine (TMR) (Invitrogen), Alexa 594 (Invitrogen) or Cy5 (GE Amersham). We used TMR for the probes against end-1 mRNA, Alexa 594 for end-3 mRNA and Cy5 for elt-2 and med-1,2. The only exceptions to this labeling scheme are for the skn-1 and pop-1 experiments in the supplement; the fluorophores used in those experiments are described in the figure legends. Optimal probe concentrations during hybridization were determined empirically. Imaging involved taking stacks of images spaced 0.3 microns apart using filters appropriate for DAPI, TMR, Alexa 594 and Cy5. We imaged the embryos using a Nikon TE2000 equipped with a Princeton Instruments camera and custom filter sets designed to distinguish between the different fluorophores used. During imaging, we minimized photobleaching through the use of an oxygen-scavenging solution utilizing glucose oxidase¹⁶. We also reduced the out of focus light by squeezing the embryos between two coverslips, thus reducing their extent in the z direction.

Image analysis

We segmented the embryos manually and manually counted the nuclei in each embryo with the aid of custom software written in MATLAB (Mathworks, Natick MA). We discounted all embryos with greater than 200 nuclei from our analysis because such embryos are developed to the point where the key developmental decisions have already been made.

We counted the number of fluorescent spots, each of which corresponds to an individual mRNA, using the semi-automated method described in ref. 16. We estimate our mRNA counts to be accurate to within 10–20%. We also counted the number of cells expressing either end-1 or elt-2 by manually selecting positive cells in every embryo.